AI World Executive Summit recap: How Enterprises Need to Look at AI


Lux Research recently helped co-chair the Artificial Intelligence (AI) World Executive Summit in San Francisco, CA – During the first day alone, we had the opportunity to hear from a diverse set of speakers, each esteemed in their own right, on their thoughts on artificial intelligence and how some of them were applying these technologies in their respective fields. Two sessions, in particular, did a fantastic job at providing two distinct perspectives on artificial intelligence that serve as sound guidelines for how enterprises should approach the hype surrounding the field. They are each summarized below:

AI, Automation, and Occupations — What The Future Could Hold

The first speaker that we heard from at the conference was Michael Chui, a partner at the McKinsey Global Institute. Michael explained how he and his team wanted to get a deeper understanding of how AI would influence the labor force beyond the simple scope of job replaceability. Since almost all current AI technologies are narrow in their application scope, he and his team wanted to analyze the specific tasks in any given occupation that could be replaced by AI technology.

To do this, the team compiled all of the occupations analyzed by the Bureau of Labor Statistics and segmented each occupation into a number of individual activities that were needed to effectively carry out the function of the occupation. Each of these activities were then mapped onto the capabilities needed to execute the activity; the capabilities were classified as being either social (e.g. social & emotional reasoning), cognitive (e.g.
understanding natural language), or physical (e.g. mobility). Since the team knew the degree of automatability for each of the capabilities mapped out, the team was able to use this segmentation to estimate the degree of automatibility for each of the occupations considered.

They predicted that 45% of the 2000 activities analyzed that individuals are paid to perform can be automated by currently available technologies and that 60% of occupations could have 30% or more of their constituent activities automated. However, while many activities are automatable, the group still predicts that fewer than 5 percent of occupations can be entirely automated using current technology.

This, Michael claims, suggests that the true value in AI technology will not be realized from a labor cost savings perspective; rather, AI will simply free up time for laborers to tackle more complex tasks and ultimately allow them to achieve a higher production throughput.

2016 and Beyond — The Rebirth of AI

We next heard from a panel consisting of Beena Ammanath, the Vice President of Data and Analytics at General Electric; Dr. Neil Eklund, a Chief Data Scientist at Schlumberger; and Kumar Srivastava, the Vice President of Product & Strategy at the Bank of New York. When asked about the burgeoning field of artificial intelligence, all three of the panelists were quick to note that artificial intelligence has been undergoing somewhat of a cycle of heightened interest over the past two decades. However, the new availability of large amounts of digitized data coupled with the scalable hardware infrastructure enabled by distributed computing has made this iteration of the AI cycle especially noteworthy since engineers
now have the tools in place to make some of this hype a reality.

Be it as it may, Beena still emphasized that the utility of artificial intelligence should not be overemphasized and that enterprises should not seek to use deep learning just for the sake of employing the new technology. She explained that enterprises should first start by defining their problem scope and then formulating a solution strategy to tackle this problem, which may or may not incorporate AI techniques when it is all said and done. Kumar and Neil both strongly echoed this sentiment, and Kumar brought up yet another interesting point. He stated that if a team forcibly incorporates AI into the solution strategy, then they may find themselves squandering time and money on the maintenance of the system for what may turn out to be a low-value use case.

Neil also brought up the often-overlooked key idea that any AI model using a supervised learning technique (the most popular in industry today) inherently requires massive amounts of quality data to perform well; if a business does not have quality data on file, then AI may not even be feasible.

Moreover, the panelists unanimously agreed that any AI as a service platforms offered by third parties should be approached cautiously, if approached at all, since any AI system requires someone familiar with the in-house data to perform feature engineering and tune the system parameters to ensure that the model is behaving properly. Rather than simply contracting out third parties, Kumar explained that enterprises should treat AI much like IT services and try to devise long-term strategies to incorporate AI into their business

A Luxonian’s Perspective:

Lux Research has been advocating much of the sentiments in the sessions discussed above for quite some time now (see the September 25, 2015 LRASJ). AI has the capacity to dramatically increase production capacity if applied correctly. If applied incorrectly, though, what was originally meant to increase process workflow efficiency can end up being the source an enterprise’s squandered time and money. This is why it is essential for businesses to understand that AI should be treated as a tool; and, as with any tool, it should only be employed when the problem solution naturally prompts an AI approach. However, to
prepare for the potential need for an AI-centric solution, enterprises should make sure that all of their data is not being stored in inaccessible silos across their business and that the data being stored is consistent and of high quality; this will make the data preprocessing step involved with training any AI model much more manageable and lead to faster turnaround.

From an executive’s perspective, then, if AI is to be thought of as a tool, then organizations should learn to effectively use this tool. Those that fail to do so risk being outpaced by competitors who are choosing to adapt to the changing pace of technology and are
incorporating AI into the fabric of their organization to help maximize their production capacity.

by Ahmed Khalil, Lux Research, Inc.